Repurposing Image Diffusion Models for Adversarial Synthetic Structured Data: A Case Study of Ground Truth Drift
Adam Arthur, Christopher Schwartz · 2026
Read Repurposing Image Diffusion Models for Adversarial Synthetic Structured Data: A Case Study of Ground Truth Drift by Adam Arthur, Christopher Schwartz (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-00788.txt.
Description
Adam Arthur and Christopher Schwartz test whether a public image generator can be misused to fake structured data. They reshape each row of the UCI Adult Income dataset into a small single-channel pseudo-image and feed it to an unmodified Stable Diffusion model. Because the model expects nearby pixels to relate, where each feature is placed in the layout becomes a key design choice.
Beyond the test, the authors draw two distinctions. They separate statistical realism, checked by machines, from perceptual realism, judged by human senses. They also define synthetic evidence, made for machines in closed pipelines rather than for people. Success there can cause ground truth drift, where fake outputs are later treated as real.
Catalog record
| CREATOR | Adam Arthur, Christopher Schwartz |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.00788